pullrequests Search Results · bucket language:JavaScript language:JavaScript language:Python language:Java language:JavaScript
Filter by
1M results
Decode graphs were captured lazily per (bucket, generation), so the first think request paid for its own captures while
serving -- one per bucket boundary its sequence crossed (512 and 1024 for a ...
diffusion
enhancement
high priority
shihongzhi
... .*, which SGLang remaps only after each bucket arrives. Miles therefore split wq_a + wkv or
compressor.wkv + wgate across update calls, and SGLang asserted on a half-filled fusion cache.
Validation ...
bobboli
... , private bucket under the same key. If the
copy fails the photo is left untouched rather than overwritten, so the original can never be lost. The copy is skipped
if the master is already there, keeping re ...
afrieirham
Summary
- extend the shared prefill CUDA Graph ladder with 1- and 2-token buckets
- centralize effective capture-cap and bucket resolution across model defaults and operator overrides ...
AkazaAkane
ayam04SUMMARY
Grained Drill by carries the clicked temporal bucket as a grained equality, like Drill to detail, while preserving the
chart s existing time filters and effective dashboard grain. This ...
doc
i18n:brazilian
i18n:chinese
i18n:czech
i18n:dutch
i18n:french
i18n:italian
i18n:japanese
i18n:korean
i18n:latvian
i18n:persian
i18n:portuguese
i18n:russian
i18n:slovak
i18n:spanish
i18n:traditional-chinese
i18n:ukrainian
i18n
packages
plugins
risk:chart-metadata
size/XXL
aminghadersohi
... edge cases (growthbook#7346)
- rollouts using n coverage (growthbook#7425)
- bucket range-end cases (growthbook#7375)
- server-side payload reduction (growthbook#7238)
- payload reduction ...
andywhite37
jadedlebi... the nearest captured bucket (3000 → 2816, 4500
→ 4096, 4607 → 4096). The decode path appends the ceiling when it is off-grid, and the CPU graph path unions it in;
prefill was the odd one out. Since ...
he-yufeng
... graph per sequence bucket. Each row is right-padded to the smallest
bucket that holds it, which is exact because every layer is causal. The LoRA adapter is merged in fp32 at load. The
shared ...
anwschne